ResearchBenchmarksScientific Reasoning on ResearchQA (test)Follow73.1AccuracyRCSD62.80465.47768.1570.823Jun 17, 2026Evaluation ResultsMethodMethodLinksAccuracyRCSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0673.1OPSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0672.8Rubric-GRPOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0672.1GRPOBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=5002026.0671.5Qwen3-8BBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0664.9SFTBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0663.2